Regular vs irregular

LABO splits every capture stream into one of two families based on when each row is written. This page is the framework the rest of the Data Streams docs build on. Start here; everything else assumes you know which kind of stream you're looking at.

The idea

• Regular streams emit a row on a fixed cadence — once per LateUpdate (render frame) or once per FixedUpdate (physics tick). The row rate is determined by the engine's loop, not by what's happening in the experiment. Even if nothing interesting occurred this frame, a row is still written.
• Irregular streams emit a row only when something specific happens — a variable is updated, an expression is detected, a button is pressed, an agent takes a step. No trigger, no row. The rate is data-driven.

Regular streams

One row per frame or physics tick. Used for continuous signals that always have a value — position, rotation, gaze direction, skeletal pose.

Stream	File	What it captures
ExperienceState	ExperienceState.csv	Authoritative per-frame timeline. Join key for every other file.
Participant	ParticipantTracking.csv	Camera pose, controller pose + velocity, grounded flag.
Body	BodyTracking.csv	Full-body skeleton joints.
Face	FaceTracking.csv	Face-mesh bones / blend-shape-driven skeleton.
Hand	LeftHandTracking.csv, RightHandTracking.csv	Left and right hand skeletons (two separate files).
Eye	EyeTracking.csv	Center / left / right eye positions, directions, gaze data.

All regular streams hook LateUpdate by default. They sample the simulation state at the end of the frame, before rendering.

In action

At 90 Hz, each regular stream produces ~90 rows per second — consistently, whether the participant is moving or sitting still. Ninety rows looks the same whether a stimulus just appeared or the screen is blank.

Irregular streams

One row per trigger. Used for discrete observations that only exist at specific moments — a variable assignment, a detected facial expression, a key press, an agent reward.

Stream	File	Trigger
Events	EventPoints.csv	Event-point firing, action execution, epoch transitions.
Variables	Variables.csv	Every variable write (assignment, modification, list update).
Expressions	Expressions.csv	Facial / hand expression detections above threshold.
Agents	AgentsData.csv	Agent observations, actions, rewards, episode state.
Input	InputCapture.csv	Key down / up / hold events for configured keys.
Interactives	(see page)	Interactive object state changes.

Each stream's writer exposes a public RecordEntry(observation) method that external code calls when the trigger fires. The observation is stamped with the current frame / time / epoch state at that instant via LaboObservation.Stamp(), then queued for the writer.

In action

During a one-minute session a participant presses the spacebar 12 times. InputCapture.csv has 12 rows, one per press event. ParticipantTracking.csv, a regular stream, has ~5400 rows (90 Hz × 60 s) — the 12 button presses are not visible as distinct events in it, only as sudden changes in controller velocity or position.

How to tell which one you need

I want to...	Use a...
Track a continuous signal over time (position, gaze, pose)	Regular stream
Count discrete occurrences (presses, detections, events)	Irregular stream
Align two continuous signals	Two regular streams joined on FrameNumber
Correlate a discrete event with continuous context	One irregular + one regular, joined on FrameNumber
Compute a rate (events per second)	Irregular for the numerator, session duration for denominator
Compute an average (mean gaze, mean velocity)	Regular, grouped by epoch / condition

Shared columns

Every CSV — regular or irregular — starts with the same LaboExperienceState prefix, so any two files can be joined on FrameNumber or MonotonicExecutionTime. Each stream then appends its own domain-specific columns after the shared prefix.

Shared columns, in order:

ObservationTime, MonotonicExecutionTime, FixedIntervalNumber, FrameNumber,
EventNumber, EpochNumber, EpochName, FrameTime, FixedIntervalTime

See Column conventions for the full table and semantics.

Terminology

• ✅ "regular" / "irregular" — the categories on this page.
• ✅ "trigger" / "the trigger fires" — what causes an irregular row to be emitted.
• ❌ "event-driven" — ambiguous with EventPoints.csv / EventNumber; don't use.
• ❌ "behavioural data" — historical grouping that confusingly excluded ExperienceState / Participant. Don't use.

Further reading

• Timing — when exactly each row is stamped.
• Join keys — how to align rows across streams.
• Column conventions — the shared prefix in detail.